Kelly Re-Visited

[Bsims]

Why do you need to demonstrate something that has already been proven? .....

Tsty made several good points in this post and raised a couple of questions. First, I’d like to answer this question. In my first post I indicated the purpose was to resolve the issue of for me. I was interested in using Kelly for my own purposes, but when used in the past I would have had better success flat betting the same amount. I believe it was user u21c3f6 that pointed out that the reason that I noticed this in the past was that my expected return was likely inaccurate. That’s probably so, since I do believe it is impossible for anyone to have an accurate expected return for a game occurring in the future.

User KVB had the one message I think we should all take away from this discussion, “It really does depend on your goals and time frame”. My goal is entertainment with a potential profit and not risking a major loss. This is a difficult goal to quantify. The time frame is easier. I’m 73 and in poor health. Each season is potentially my last. So, my objective is to show a profit of whatever size. Making a lot of money at the end of my life isn’t important. I’ve already got that covered for my wife.

Tsty also raised the issue of bet sizing that most bettors don’t pay enough attention to. Picking the side is obviously the most important decision to make. If you can’t pick wagers with a +EV, then no bet sizing will help. However, you can pick winners with +EV but still be a loser because of money management. Tsty raised his game with this post.

I’ve documented my efforts in my blog. I’m about ready to summarize my thoughts for my next and last post on this issue. I’ll summarize and share with you here.

[Waterstpub87]

How is kelly criteria actually implemented?

Consider, I have a 10,000 bankroll, and I have the mets winning a game 50% of the time, and the odds are currently 150. Would a full kelly be 1,000 bet, a half 500 bet, a quarter 250?

[Bsims]

How is kelly criteria actually implemented?

Consider, I have a 10,000 bankroll, and I have the mets winning a game 50% of the time, and the odds are currently 150. Would a full kelly be 1,000 bet, a half 500 bet, a quarter 250?

In your example the odds are 2.50 and estimate is 0.50.

Kelly % = (Odds*Estimate-1)/(Odds-1)
Kelly % = (2.50*0.50-1)/(2.50-1) = (1.25-1)/1.5 = .25/1.5 =0.1666...

Full Kelly = $1666.67 Half Kelly = $833.33 Quarter Kelly = $416.67

[Waterstpub87]

In your example the odds are 2.50 and estimate is 0.50.

Kelly % = (Odds*Estimate-1)/(Odds-1)
Kelly % = (2.50*0.50-1)/(2.50-1) = (1.25-1)/1.5 = .25/1.5 =0.1666...

Full Kelly = $1666.67 Half Kelly = $833.33 Quarter Kelly = $416.67

Thanks for this. Seems like a pretty crazy high amount to be on stuff.

[LT Profits]

Thanks for this. Seems like a pretty crazy high amount to be on stuff.

Probably because 50% probability on a +150 is unrealistic. Any model that is off by that much very often is usually useless.

[Waterstpub87]

Probably because 50% probability on a +150 is unrealistic. Any model that is off by that much very often is usually useless.

How much should a good model be off, considering a selection rate about 20% or so?

[Rich Boy]

How much should a good model be off, considering a selection rate about 20% or so?

Depends on the market, you can expect greater inefficiency in smaller markets, so I would target those (ie. Props markets, totals, smaller leagues). If you focus on the markets the books dont care about you are more likely to catch off numbers. Overall I think sports markets have become increasingly efficient and it is harder to find edges than it used to be.

[ericc]

There have been some recent posts dealing with flat betting versus varying the bet size based on the expected returns (i.e. Kelly Criterion). I’d like to begin an intelligent discussion of the two alternatives. We all have opinions and I’d like for us to share those. But, please provide some documentary support for your opinions.

I really want to believe in Kelly and have used it in the past (and am currently using a variation). Unfortunately, my past experiences have cast doubt on my feelings. I’ve decided to put in a fair amount of effort to finally resolve the issue for me. I will be documenting this project in detail on my blog, ole44bill.blogspot.com. I really want to encourage discussion here on what I find, both supporting and non-supporting.

Since I think most people here are anti-flat betting, here is a link to an anti-Kelly discussion.

http://www.professionalgambler.com/debunking.html

If you have thoughts on this guys article, please share them.

One of the things he mentions is that each time you 'jiggle' the size of your bet, the break-even figure that the gambler must surpass, increases.

Any input on that?

[u21c3f6]

I’m 73 and in poor health. Each season is potentially my last. So, my objective is to show a profit of whatever size. Making a lot of money at the end of my life isn’t important. I’ve already got that covered for my wife.

Bsims, I am sorry to hear of your poor health. Hopefully it is not too impactful on your daily life.

If your objective is to show a profit in a relatively short term, then my suggestion is to steer clear of Kelly. First, in order to use Kelly, one must be profitable. If one cannot show a profit on flat wagers, then you cannot use Kelly because the Kelly wager for a 0 or negative expectation is $0.00.

Kelly is best used for the longer term with relatively close estimates of your edge. No edge, no Kelly wager. Don't know your edge, then what would you input into the Kelly formula to determine wager size?

Good luck.

Joe.

[Waterstpub87]

Depends on the market, you can expect greater inefficiency in smaller markets, so I would target those (ie. Props markets, totals, smaller leagues). If you focus on the markets the books dont care about you are more likely to catch off numbers. Overall I think sports markets have become increasingly efficient and it is harder to find edges than it used to be.

Bet more props than anything. Up over 25 units in MLB props this season. Doesn't particularly answer the question of how off should a good model find MLB sides to be.

[Bsims]

I finally got around to posting my conclusions resulting from my Kelly simulations on ole44bill.blogspot.com. Following are the main points.

• · Kelly has the potential to substantially outgrow flat bets over a long period of time. (Advantage Kelly.)
• · Kelly also limits your losses to a portion of your bankroll. (Advantage Kelly).
• · It is also better than a similar system that I devised. It may very will be the optimum approach of maximizing final return (Advantage Kelly.)
• · Kelly is dependent on how accurately you can estimate your edge in each wager. (Disadvantage Kelly.)
• · If you are dealing with each season independently, you have a limited number of potential bets. Kelly is superior with an unlimited number of future bets. (Disadvantage Kelly).
• · Kelly is more fun because it introduces another unknown gamble. (Advantage Kelly.)

Again I pointed to KVB’s quote, “It really does depend on your goals and time frame”. Because of my time frame (the rest of the 2018 MLB season) I’ll be sticking with my current wagering approach for my primary betting.

I have decided to implement a Kelly based scheme for a new system I’m experimenting with for fun. The bankroll was started with about half of my average bets for one day. I’ll probably also carry this approach into the football season.

[tsty]

Probably because 50% probability on a +150 is unrealistic. Any model that is off by that much very often is usually useless.

Why?

Just look at how much lines move every day and you can see why you are wrong. Heck even in the NBA moving a few points is not out of the question.

[tsty]

How much should a good model be off, considering a selection rate about 20% or so?

Your aim should not be to make a model that is different from the market but a model that is accurate. Everything else comes after that because if you are accurate then you will always make money.

On a site note I would say it's impossible for any model to be beating pinny closing lines at 20% otherwise you would be bankrupting every syndicate in existence.
In all honesty thou I don't think anyone going at it alone should be in any major markets. It's basically impossible to be making any meaningful money in NBA MLB NFL etc. Just too much competition out there.

[Bsims]

I've always computed expected return per dollar as output from any model I make. While back testing, I group expected returns like $1.00 to $ 1.05, $1.05 to $1.10, etc. Then computed the actual returns and plot them. I've noticed as the expected returns get higher the actual returns start to fall off, kind of in a bell shape. At the extremes, they are worthless.

So I call this modeling a game that is different from the one being played today. For instance, a star player might be unavailable and the model wasn't aware of that. But the point of where to be wary varies with the model and isn't really easy to determine.

(By the way, another negative for Kelly if you blindly follow a model's projection).

[Waterstpub87]

Your aim should not be to make a model that is different from the market but a model that is accurate. Everything else comes after that because if you are accurate then you will always make money.

On a site note I would say it's impossible for any model to be beating pinny closing lines at 20% otherwise you would be bankrupting every syndicate in existence.
In all honesty thou I don't think anyone going at it alone should be in any major markets. It's basically impossible to be making any meaningful money in NBA MLB NFL etc. Just too much competition out there.

I think you misunderstand. A selection rate of 20% means finding around 1/5 of games to be off enough to be playable. Not beating the close by 20%

[HeeeHAWWWW]

I've noticed as the expected returns get higher the actual returns start to fall off, kind of in a bell shape. At the extremes, they are worthless.

Yep indeed, very familiar with this one.

You can adjust to get a truer EV estimate to feed into Kelly, but that has some dangers. I prefer just to not bet above a specific EV cutoff, since it clearly implies the market knows something I don't.

[LT Profits]

Why?

Just look at how much lines move every day and you can see why you are wrong. Heck even in the NBA moving a few points is not out of the question.

We are talking about a 10% variance in a major market here though. How often do MLB lines move by at least 10% implied probability? And also, I did say "off by that much very often", a few outliers here and there is fine but if model shows a lot of 10% variances, it is probably junk.

[Rich Boy]

We are talking about a 10% variance in a major market here though. How often do MLB lines move by at least 10% implied probability? And also, I did say "off by that much very often", a few outliers here and there is fine but if model shows a lot of 10% variances, it is probably junk.

Agreed. Opening to closing numbers in any major market are never off by more than a few percent on average.

I love when people, specifically touts, claim to hit 60%+ on NFL sides. Any person or any model that claims numbers like that are frauds. The lines are too efficient.

Trying to beat NFL sides is like trying to beat forex, nearly impossible.
Why not go for the penny stocks of sports like props.

Most handicappers never give themselves a chance at this game trying to tackle the most efficient and liquid markets out there (NFL, NBA sides). But then again most people do this for entertainment.

[Rich Boy]

Here is a tip for any rookies out there:

Look through every line offered by your book and find the ones with the lowest limits, start there.
Find a model that can handicap these lines better than the market.

A perfect example is any prop bet, CFL, WNBA.

Limits are low on these markets for a reason, books arent that confident in the efficiency of lines. Meaning you have a chance of handicapping the lines better than them.

Never bet major markets unless you have inside info, dont even fkn try it. NFL, forget it. NBA, forget it.

Do that and you mind make it, but probably not.

[Bsims]

When you start a project you have some expectations of what you will find. But it doesn't usually work out that way. I wanted to find a bet sizing strategy that works for me. Well, I kinda have gotten there. In my blog post from Monday I have a revealing chart (for me) that lead me to conclude that I would pursue a star-based approach for the rest of the MLB season and forward. (The star-based system I described is a cross between Kelly and flat betting.) But when I went to implement the star-based approach, it became obvious that it was not practical for my primary MLB system and that continued flat betting was the best way to continue. But I had been using a star-based system for my secondary system and based on my analysis, I'm switching that one to a Kelly approach. I'll be discussing the reasoning in my next post. Funny how things work out. It's time to put this topic to sleep and to move on to other efforts.

[Bsims]

Yep indeed, very familiar with this one.

You can adjust to get a truer EV estimate to feed into Kelly, but that has some dangers. I prefer just to not bet above a specific EV cutoff, since it clearly implies the market knows something I don't.

I n my latest post, ole44bill.blogspot.com, I have a couple of charts showing a well behaved system and one not so well behaved. As HeeeHAWWWW said, it's quite a familiar problem.

[Miz]

I can't read the whole thread but I'll offer this from real experience actually executing the same model for 6 seasons now. I've bet nearly full kelly for an entire season. The swings were enormous, and I probably overbet at times. I was also a little bit lucky (large bets won more than the small ones) and it worked out. I also bet half kelly and the way it worked out I lost my big bets (days with very few plays) and won the smaller ones... went 54.5% and lost money. It comes down to how well you quantify your edge (filter out crazy plays), and how you fare psychologically with large swings.

I have gone back to a slightly more conservative approach because it appeals to me. It is hard to know exactly what your edge is, so I use about 0.33 to 0.5 kelly to account for the slop.

Best of luck with your health Bsims. Take care.

[trytrytry]

great thread

[danshan11]

It really does depend on your goals and time frame but I agree that changing your bet size as least as possible is a solid strategy for sports betting.

Nobody survives going full kelly and those that say they do are really only fractional.

Changing your bet size changes your break even, and usually not for the better.

I've been making posts like that link since I came to SBR and you're right, it has been controversial.

Goals matter, you running a business that needs to draw income? Long term growth? For fun? Goals matter, as with any sound financial plan.

All the theoretical math in the world won't change that.

it is impossible to truly use kelly in sports betting because we do not know our edge until the game closes and that is too late. betting your average edge is basically flat betting, only thing that resembles a smart kelly play is fixed profit betting bet 1.X to win 1 faves and bet .x to win 1 dogs,

[u21c3f6]

it is impossible to truly use kelly in sports betting because we do not know our edge until the game closes and that is too late.

Depends on how one calculates their edge. I think most try to calculate their edge for each individual wager by using the closing line but I don't do it that way. If I told you that on average I cash 53% of my -105 wagers can you tell me what my edge is for my next wager before the game closes? That is how I do it.

Joe.

[HeeeHAWWWW]

it is impossible to truly use kelly in sports betting because we do not know our edge until the game closes and that is too late.

This is a common myth - you don't need to know your edge exactly, merely have an estimate that is on average correct.

[danshan11]

Depends on how one calculates their edge. I think most try to calculate their edge for each individual wager by using the closing line but I don't do it that way. If I told you that on average I cash 53% of my -105 wagers can you tell me what my edge is for my next wager before the game closes? That is how I do it.

Joe.

good way to do it but that is essentially flat betting with a twist for when you are winning more, I dont trust that for stability. I truly believe in fixed profit betting, risk more and more likely to win games and less on less likely to win games and always playing to 1 unit.

[danshan11]

This is a common myth - you don't need to know your edge exactly, merely have an estimate that is on average correct.

if you estimate your edge, you are essentially flat betting, which flat betting is not bad but just try it go back and change your history to fixed profit betting and see how you would have done. My guess is more times than not you would make more, its not a huge difference but I common sense wise like the approach. bet more on more likely wins and less on less likely wins, the issue here is the change of my CLV unfortunately my best clvs are dogs and with me betting less on them this lowers my overall clv if you weigh it.

[danshan11]

I can't read the whole thread but I'll offer this from real experience actually executing the same model for 6 seasons now. I've bet nearly full kelly for an entire season. The swings were enormous, and I probably overbet at times. I was also a little bit lucky (large bets won more than the small ones) and it worked out. I also bet half kelly and the way it worked out I lost my big bets (days with very few plays) and won the smaller ones... went 54.5% and lost money. It comes down to how well you quantify your edge (filter out crazy plays), and how you fare psychologically with large swings.

I have gone back to a slightly more conservative approach because it appeals to me. It is hard to know exactly what your edge is, so I use about 0.33 to 0.5 kelly to account for the slop.

Best of luck with your health Bsims. Take care.

I curiously ask, how do you know your edge or imply your edge?

[u21c3f6]

good way to do it but that is essentially flat betting with a twist for when you are winning more, I dont trust that for stability. I truly believe in fixed profit betting, risk more and more likely to win games and less on less likely to win games and always playing to 1 unit.

Not sure how you came to the conclusion that what I wrote (and what HH wrote) was essentially flat betting. It is not clear to me that you understand Kelly wagering.

Joe.

[danshan11]

if you use kelly to calculate your edge you need to know your edge so if your edge is X you pop X into kelly every single time you bet so X is not a factor anymore since it stays basically the same and that is flat betting, I could be wrong been wrong many times before, so no issue there but please clarify where I am missing what you are doing because I feel like my understanding of kelly is fairly sound, please advise thanks again!

[danshan11]

if you are using your probabiltiy based on bet at line , you are fixed profit betting and I like that. but to clarify how are you calculating your implied?

[danshan11]

you either have edge or win probability
win prob is fixed profit betting
edge is impossible unless you know the closing line which we dont
edge is based on your edge over a zillion games and that is flat betting\
right?

[danshan11]

also if you cash 53% of your -105 tickets it dont matter how you stake other than all in or martined you will win!

[u21c3f6]

if you use kelly to calculate your edge you need to know your edge so if your edge is X you pop X into kelly every single time you bet so X is not a factor anymore since it stays basically the same and that is flat betting, I could be wrong been wrong many times before, so no issue there but please clarify where I am missing what you are doing because I feel like my understanding of kelly is fairly sound, please advise thanks again!

You use Kelly to calculate your wager size not your edge. First you must have an edge and at least a close estimate of what that edge is before you can use Kelly. Then your wager size as a % of your bankroll is determined by Kelly as your edge divided by the odds. The actual dollar amount of your wagers will fluctuate with changing edges, odds and/or bankroll size.

Joe.